Terence Collier, Bal Benavidez, and Indira Gubeljic
The last few steps in wafer fabrication are critical for high reliability interconnection. Those steps, where the bond pad final metal is deposited and etched, and the wafer passivation is added and etched to define the pads, also define the long-term reliability of the later connections to that pad by stud bump or wire bond.
For typical aluminum bond pads, the passivation layer of nitride, oxide, or oxynitride is dry-etched with a fluorinated gas plasma containing either sulphur hexafluoride (SF6) or carbon tetrafluoride (CF4) These fluorinated plasmas selectively etch the oxide or nitride passivation, but remove little of the underlying aluminum pad.
Unfortunately, after dry etching some fluorine remains in the aluminum grain boundaries and interstitial sites, or forms thin layers of AlFx, typically AlF3, on the surface. Thin layers of aluminum oxides and hydrates on the surface may also be converted to aluminum oxyfluoride (AlOF) by the plasma.
The residual fluorine acts as a catalyst in the presence of water, so that hydrates and hydroxides on the surface form hydrofluoric acid, etching the aluminum and creating additional hydrates, oxides, and oxyfluorides.
The corrosion layer can grow to 600 angstroms thick, compared with native aluminum oxide, which self-limits to 80 angstroms. Figure 1 shows a cross-section of a wire bonds with no cleaning, after 500 hours at 150C.
Figure 1. Wire bond intermetallic layer after 500 hours at 150C. (IEEE 2006)
The thick corrosion layer makes resistance probing difficult. It interferes with wire bonding and stud bumping, since the gold ball of the bonder must pierce the corrosion layer and form a gold-aluminum intermetallic bond at the aluminum surface. A thick corrosion layer might even trap some of the corrosive material at the bond interface.
The corrosion layer must be removed to improve bumping, bonding, and probing. Argon plasma ashing has varying success in removal, but layer thicknesses and composition differ across the wafer, and the aluminum surface may still contain interstitial fluorine that can become embedded in the later wire bond or stud bump joint.
Two methods show improvement over argon ashing in removing all of the fluorine, oxides, and corrosion products. One method is to clean the etched surface with a liquid chemical remover. The other is to replace the aluminum surface with bondable layer of gold or palladium metal. Both will be discussed and compared below.
Air Products BPS100 and BPS101 are liquid solutions that remove the oxide, oxyfluoride, and hydrated layers without attacking the aluminum. The solutions quickly remove the corrosion layers, typically within 5 minutes at room temperature.
The second method is to substitute a bondable layer of gold or palladium for the aluminum bond pad. This avoids Au-Al intermetallics, as well as aluminum oxides, fluorides, or oxyfluorides in the bond.
A common approach is the electroless nickel – immersion gold process, which replaces a thin layer of aluminum with 1µm to 2µm of nickel and 0.05µm to 0.06µm of gold.
For higher temperature longevity, a barrier layer of 0.2µm palladium may be added between the nickel and the gold to prevent nickel from diffusing into the gold. Another alternative formulation we tested is Ni(Pd/Co)Au.
To compare these and other approaches we used a standard 5x5mm ASIC die taken from a 200mm silicon wafer, with 80µm by 80µm aluminum bond pad openings in Silicon Nitride (Si3N4 ) passivation.
Aluminum thickness was a nominal 0.8 µm. The sawn die were stored at ambient temperature to accelerate corrosion.Test groups were made up by randomly selecting die from the saw tape. The groups were:
- No cleaning (control group).
- BPS100 cleaning
- NiAu (thicker AU)
All samples were gold stud bumped with a K&S manual bonder at 150°C. Ageing as described below was in a convection oven at 150°C. The testing sequence for all die was :
1. No aging – first set of pads bumped within 2 hours of removal from plating.
2. Two hours—second set of pads bumped.
3. Four hours – third set of pads bumped.
4. Six hours – fourth set of pads bumped, after 1.5 hours on the wire bonder at 150; total 7.5 hours.
The first data collected is bondability, that is, how well did the bump adhere in bonding? Table 1 shows these qualitative bondability results for different processes and times.
|Split||Time zero||2hrs @150C||4hrs @150C||6hrs @150C|
|Al||Some not sticks||Zero sticks||Zero sticks||Zero sticks|
|Al +BPS100||Good||Good||Some no sticks||More no sticks|
Table 1. Wire bondability based on pad metallization
After bonding, samples were potted and polished for further analysis, including wire pull, ball shear, cross-sectioning and Auger Depth Profile analyses.
After heating and processing, some flash gold layers show lessened quality from time zero. Analytical testing shows that nickel bleeds through the thin gold and then oxidizes, increasing the resistance and decreasing the bondability. Adding palladium solves this problem.
In this view, gold and palladium compliment one another. The palladium prevents nickel from diffusing through the gold, and the gold protects the palladium from contamination. The bond or bump forms a stronger joint with gold and with palladium.
Test units were also evaluated for bondability after exposure to room temperature and humidity (25°C and 65% relative humidity) for varying times. Table 2 shows the comparison. Again, all treated units performed better than untreated aluminum pads.
|Wire bondable t=0||Wire bondable t=24hrs||Wire bondable t=72|
|Cleaned aluminum +DI water soak||No||No||No|
Table 2. Wire bondability after soaking at ambient.
Some tests of gold-to-gold contacts at 200°C showed that the advantages of gold to gold contacts continue well beyond the standard 150°C high reliability standards. This could allow gold to gold replacement of more exotic and expensive materials in high temperature semiconductors.
Wire pull, die shear, and Auger results are presented in detail in our original paper.
- Removing the corrosion layer on the surface of aluminum bond pads greatly increases bump and bond reliability.
- This layer forms in passivation etching, and continues for the life of the device if moisture is present.
- Best results occur when a layer of NiPdAu, Ni(PdCo)Au, or NiAu is substituted at the bond interface.
- These added layers also improve manufacturability and yield, at a cost of a few cents per die.
- Improved performance of gold to gold contacts may extend to 200° and beyond.
FOR MORE INFORMATION
T. Collier, B. Benavidez,I. Gubeljic, “High Reliability Bond Pads,” Proceedings of SMTA International Symposium 2010, page 399